A Novel Methodology to Recognize Drivers’ Distraction using Support Vector Machine Classifier

Abstract

Monitoring driver fatigue, inattention, and lack of sleep is very important in preventing motor vehicles accidents. An efficient system for automatic driver vigilance should make use of physiological, behavioral and car measurements. The driver distraction system is often performed by supervised classifiers, which require an adequate amount of labeled instances to train the classifier. All of these classifiers depend upon the quality and quantity of the training set used to train the classifier, whose reliability is a fundamental issue for an accurate mapping of the investigated area. Support Vector Machines (SVMs) are one of the successful classifier applied for driver inattention detection applications. SVMs are nonparametric statistical approaches for addressing supervised classification and regression problems. Therefore, there is no assumption made on the underlying data distribution. In the original formulation of SVMs, the method is presented with a set of data samples, and the SVM training algorithm aims to determine a hyperplane that linearly divides the data set into two classes. The term optimal separating hyperplane is used to refer to the decision boundary that minimizes misclassification attained during the training phase. Learning refers to finding an optimal decision boundary to separate the training patterns and then to separate test data under the same configuration. The crucial part for any kernel-based technique, including SVMs, is the proper definition of a kernel function that accurately reflects the similarity among samples. This research work addresses two problems of SVM, one is the kernel function selection and the other is the training time. A convex-hull and geometry based SVMs is proposed here for driver distraction detection. The proposed Convex-hull & Geometry based SVM (CG-SVM) doesn't require a kernel function and a training time too. In this way, the complexity of SVM is much reduced while preserving the driver distraction detection accuracy. The performance of the proposed CG-SVM is studied with the Ford challenge driver distraction database received from 2011 International Joint Conference on Neural Networks.